Five Checks Before You Import an AI Site Model into OpenRoads

Before importing an AI-generated site model into OpenRoads, it is crucial to conduct a series of checks to ensure the model is accurate and usable for design purposes. The five key checks are:

1. Coordinate system and units: Verify the file format and coordinate system used in the export. OpenRoads requires specific units, such as feet or meters. Measure a road width visible in imagery to confirm the scale. If the imported model appears plausible but is significantly off, there may be a units issue.

2. Vertical datum and elevation range: Check the minimum and maximum elevations against a trustworthy benchmark. Ensure the site's elevation is not significantly different from expected values. A constant difference can be corrected, but if it varies across the site, the surface may be unusable for design.

3. Surface topology and breaklines: Remember that a mesh is not a TIN surface. It infers elevations based on triangles, which may not accurately represent sharp changes in ground direction. Look for long, skinny triangles crossing roads or walls. Turn on triangle edges to identify areas where the surface is making assumptions. Be cautious of decimated point clouds, as they can alter curb positions.

4. Scale and control: Measure two points on the ground to verify the model's scale. If the model is off by just one percent over a mile, that discrepancy can translate to 52 feet. Use GIS data only as context, not design, when checking the scale. Point cloud data decimation can also affect the accuracy of the model.

5. What the model leaves out: AI-generated site models often omit critical elements like utilities, walls, culverts, inlets, guardrails, and underground features. Ensure the model includes essential elements based on the project requirements. For example, a warehouse project may need dock heights, while a parking lot design may require drainage considerations.

Before finalizing the model, keep a small script that checks the elevation range and counts any impossible values in the exported point file. This serves as a smoke test to quickly identify obvious failures. However, remember that these checks are not validation but rather a precautionary measure to catch obvious issues before they reach the design file.

The decision to skip these checks depends on the context. For public meetings or quick visual reviews, it is acceptable to skip the entire process. However, when preparing a model for actual design work, the checks should be completed. The phrase "ready to use" should trigger the initiation of these checks. Remember that a quality check is far less costly than redesigning a flawed model.

Document the file format used and any relevant notes in the project log to maintain transparency and avoid future issues. If a model fails two or more checks, it is best to halt the process and request the original source material for re-evaluation.